Replace time.time() to time.perf_counter() for benchmarking. (#6178)

Signed-off-by: Lifu Huang <lifu.hlf@gmail.com>

benchmark/mtbench/bench_other.py

@@ -66,7 +66,7 @@ def main(args):
         answers[i] = cur_answers
 
     # Run requests
-    tic = time.time()
+    tic = time.perf_counter()
     if args.parallel == 1:
         for i in tqdm(range(len(questions))):
             get_answer(i)
@@ -79,7 +79,7 @@ def main(args):
                 )
             )
 
-    latency = time.time() - tic
+    latency = time.perf_counter() - tic
 
     print(f"#questions: {len(questions)}, Latency: {latency:.2f}")
 

benchmark/mtbench/bench_sglang.py

@@ -57,7 +57,7 @@ def main(args):
     sgl.set_default_backend(backend)
 
     # Run requests
-    tic = time.time()
+    tic = time.perf_counter()
     rets = answer_mt_bench.run_batch(
         arguments,
         temperature=0,
@@ -66,7 +66,7 @@ def main(args):
         progress_bar=True,
     )
     answers = [[s["answer_1"], s["answer_2"]] for s in rets]
-    latency = time.time() - tic
+    latency = time.perf_counter() - tic
 
     print(f"#questions: {len(questions)}, Latency: {latency:.2f}")
 

benchmark/mtbench/bench_sglang_eagle.py

@@ -68,7 +68,7 @@ def main(args):
     sgl.set_default_backend(backend)
 
     # Run requests
-    tic = time.time()
+    tic = time.perf_counter()
     rets = answer_mt_bench.run_batch(
         arguments,
         temperature=0,
@@ -78,7 +78,7 @@ def main(args):
     )
     answers = [[s["answer_1"], s["answer_2"]] for s in rets]
 
-    latency = time.time() - tic
+    latency = time.perf_counter() - tic
     num_output_tokens = sum(
         s.get_meta_info("answer_1")["completion_tokens"]
         + s.get_meta_info("answer_2")["completion_tokens"]